Uncovering Temperature-Load Sensitivity via Structure Modeling and IVGMM

Accurately capturing the temperature sensitivity of electricity demand contributes to efficiently managing power system operation. Conventional approaches typically attempt to reduce uncertainties by incorporating temperature as a feature within forecasting models to improve load forecasting accuracy. However, such models often fail to quantify the degree of temperature sensitivity at each time point. In this paper, we leverage structure modeling and Instrument Variable Generalized Method of Moments (IVGMM) to estimate the dynamic temperature sensitivity of electricity demand. Structure modeling explores the relationship between actual load value, time series forecasting results, and external weather factors through multivariate statistical analysis and Structure Equation Modeling (SEM). The time series forecasting results, representing expected electricity demand at each time point, are generated by a deep learning module including a Bidirectional Gated Recurrent Unit (BiGRU). Based on the uncovered relationships between variables, IVGMM provides robust parameter estimates to the model. In the case study, the proposed process is validated to achieve accurate quantification of temperature impact on load and well capture irregular patterns of fluctuations in load.